ReGen-DF is a hybrid deepfake face dataset that combines:
- Text-to-Image (T2I) generated faces from five modern generative models
(e.g., seedream, kling, wan, and other diffusion-based T2I models), and - Face reenactment–based forgeries built on top of FFHQ real face images using an AILab platform.
The goal of ReGen-DF is to provide a realistic and diverse benchmark for deepfake detection, face forgery analysis, and robustness evaluation under multiple generation pipelines.
The dataset is publicly available via Google Drive:
🔗 Download ReGen-DF
ReGen-DF on Google Drive
Note: Please make sure to read the License & Usage and Ethical Considerations sections before using the dataset.
-
Hybrid Generation Pipelines
- T2I-generated faces from multiple state-of-the-art models
(e.g., seedream, kling, wan, and two additional diffusion-based generators). - Reenacted faces using FFHQ real identities and face reenactment techniques (via AILab platform).
- T2I-generated faces from multiple state-of-the-art models
-
Multiple Forgery Types
- Purely synthetic faces (no real image counterpart).
- Identity-preserving reenactment forgeries based on real FFHQ faces.
-
Benchmark-Oriented Design
- Suitable for binary deepfake detection (real vs fake).
- Can be extended to source attribution (which generator produced the fake).
- Designed with research reproducibility in mind.
The ReGen-DF dataset is released under the
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.
- You may use, share, and adapt the dataset for non-commercial research purposes only.
- You must provide appropriate credit to the authors.
- If you create derivatives or adapted versions of this dataset, you must distribute them under the same license (CC BY-NC-SA 4.0 or a compatible license).
- Any commercial use is not permitted without explicit permission from the authors.
-> For the full legal text, please refer to license.txt in this repository
and the Creative Commons page:
https://creativecommons.org/licenses/by-nc-sa/4.0/
The dataset is organized as follows (example structure):
ReGen-DF/
├─ Genearaion/
│ ├─ seedream (5000)/
│ │ ├─ image_00000_sd.png
│ │ ├─ image_00001_sd.png
│ │ └─ ...
│ ├─ flux_output (979)/
│ ├─ imagen04_output (381)/
│ └─ ...
└─ Face Reenactment/
├─ output_emotions (155)/
├─ output_emotions_15000 (1480)/
├─ output_emotions_16000 (1481)/
└─ ...
| Type | Sample |
|---|---|
| T2I - seedream | ![]() |
| T2I - kling | ![]() |
| T2I - wan | ![]() |
| T2I - flux | ![]() |
| T2I - imagen | ![]() |
| T2V (basic) - kling | ![]() |
| T2V (lipsync) - kling | ![]() |
| Face Reenactment | ![]() |
If you use ReGen-DF in your research or project, please cite it as follows:
@dataset{regen_df_2025,
title = {ReGen-DF: Reenactment & Generation DeepFake Face Dataset},
author = {Jinseok Kim, Uihyeon Maeng, Heejung Jang, Suyeon Myeong},
year = {2025},
howpublished = {GitHub and Google Drive},
note = {Available at \url{https://github.com/maeng00/ReGen-DF}}
}







